v0.1 of Likewise (https://getlikewise.ai/spec/) was published on 2026-05-05. Likewise is a protocol for decentralised personal knowledge graphs: a wire-level standard that lets users own the things AI systems infer about them. There's an in-development reference implementation alongside the spec but it isn't public yet.
I'm publishing the spec before the implementation is public because the standard matters more than the app. Most of the protocols I respect (Solid, AT Protocol, the Willow Protocol) come out of teams that already had institutional credibility before they had the artefact. I don't, so I'm publishing from a different position and explicitly asking for the kind of review that institutional context would normally provide.
What I'd particularly value feedback on
- The data model. Where am I reinventing things that have already been better thought-through? Where is the evidence/claim/entity/episode shape wrong, ambiguous, or missing primitives a real implementation would want?
- The three-layer split. Substrate (Part 1), inference pipeline (Part 2), application conventions (annex). The intent is to let an organisation implement just Part 1 without the rest. Does the split read clean, or are there places where Part 1 depends on something only Part 2 specifies?
- The comparison chapter (here). Honest contrast with Solid, AT Protocol, Nostr, Iroh, Willow, and the local-first manifesto. Where is the framing unfair, or where am I missing a closer cousin?
- The conditional audit invariant (I-9). Default-on for user-run nodes, opt-in via caveat for delegated nodes. Is that the right shape, or are there scenarios where it goes wrong?
- The "isn't this just X?" critique. Please don't hold back. The comparison chapter tries to pre-empt the obvious cases; tell me where it's wrong.
How to engage
- Open a new issue for anything specific so it can be tracked individually.
- Or comment on this issue with broader reactions.
- Or email daniel@danielmay.co.uk if you'd rather not be public.
I won't promise to act on every piece of feedback, but I will read everything and reply to anything substantive. Material critiques will land in a v0.1.1 patch over the next couple of weeks. Bigger structural ones will inform v0.2.
Thanks.
v0.1 of Likewise (https://getlikewise.ai/spec/) was published on 2026-05-05. Likewise is a protocol for decentralised personal knowledge graphs: a wire-level standard that lets users own the things AI systems infer about them. There's an in-development reference implementation alongside the spec but it isn't public yet.
I'm publishing the spec before the implementation is public because the standard matters more than the app. Most of the protocols I respect (Solid, AT Protocol, the Willow Protocol) come out of teams that already had institutional credibility before they had the artefact. I don't, so I'm publishing from a different position and explicitly asking for the kind of review that institutional context would normally provide.
What I'd particularly value feedback on
How to engage
I won't promise to act on every piece of feedback, but I will read everything and reply to anything substantive. Material critiques will land in a v0.1.1 patch over the next couple of weeks. Bigger structural ones will inform v0.2.
Thanks.